菜单
Produce an in-depth, source-grounded research report. Use for broad questions requiring multiple research threads, reconciliation, and a cited report rather than a quick factual lookup.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
基于 SOC 职业分类
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/nateberkopec/dotfiles --skill deep-research命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Set up or audit Nate's standard dev environment in a project directory. Use for mise tasks, hk hooks, CI hardening, or standardized project tooling.
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
| name | deep-research |
| description | Produce an in-depth, source-grounded research report. Use for broad questions requiring multiple research threads, reconciliation, and a cited report rather than a quick factual lookup. |
Use separate agents for source gathering and final synthesis so each role keeps a clean context. The lead coordinates, interviews, delegates reconnaissance, checks coverage, and validates the final report; it does not perform independent web research.
For a narrow fact or small documentation lookup, answer directly instead of using this workflow.
Ask one focused round at a time until the audience, decision, boundaries, desired depth, source constraints, date range, output location, and success criteria are clear. Propose a research question and outline for approval. Do not begin full research before approval.
When the topic needs a methodological lens, read references/research-frameworks.md only for the applicable framework.
Launch a small reconnaissance assignment to identify terminology, source types, obvious subquestions, and likely blind spots. Use its findings to divide the outline into non-overlapping research threads. Keep a coverage map connecting every approved section to at least one assignment.
Before launching researchers or a report writer, read references/agent-briefs.md and apply the relevant contract. Give each researcher one thread and one output file under research_notes/. Parallelize independent threads; keep dependent work sequential.
Completion criterion: every outline section has an owner, boundaries, evidence standard, and note path.
Wait for all assignments, then inspect every note for source quality, missing citations, contradictions, and unfilled sections. Delegate targeted follow-ups rather than silently filling gaps. Preserve conflicting credible evidence for synthesis.
Completion criterion: every approved section has adequate primary or authoritative support, or an explicit documented limitation.
Delegate the report to a fresh writer with the approved question, outline, all note paths, and final path. Require citations near claims, explicit conflict handling, limitations, and a source list. The report must be synthesis, not pasted notes.
Review the output against the coverage map. Send specific defects back to the writer or a researcher and repeat until every section is supported and the report satisfies the agreed success criteria.
Return the report path, a concise summary of conclusions, important uncertainties, and the source count. Keep research_notes/ unless the user asks to remove it. Do not present the report as complete while a cited claim is unsupported, a credible conflict is hidden, or an approved section is missing.